Introduction
The InfraValidator component performs an early warning before pushing the model to production. The component is used only in creating Training pipelines. The Robotika InfraValidator container is the inheritor of the InfraValidator TFx [1] container with an extended UX.
Inputs
- “Operation type” is an option to choose whether to perform prediction on the test dataset and load the model or solely load the model.
- “Model tag” is a tag for the model used to identify the pipeline on the “pipeline runs” page.
- “Max loading time” is the parameter that confines the maximum loading time.
How to use
We recommend using distinct model tags for different models to differentiate among the pipeline runs.
We recommend making the max loading time at least 15 seconds. The Robotika model is based on TFX, and TFX requires additional timing for initiating the pipeline. In general, the max loading time is determined by a specific model and dataset. For example, for large datasets, the parameter can be set to 1000 seconds.
References
[1] Infravalidator TFx component, https://www.tensorflow.org/tfx/guide/infra_validator